Standard Intelligence

Standard Intelligence is an artificial intelligence research company developing general-purpose agents capable of operating computers. The company's approach deviates from traditional methods that rely on text and annotated images, instead focusing on training foundation models directly on raw video footage of computer use. This strategy involves predicting the next mouse movement, click, and keystroke from pixels, a paradigm compared to Tesla's approach to self-driving but applied to digital knowledge work.

The company was founded by Galen Mead and Devansh Pandey, who met at the Atlas Fellowship in 2022 while still teenagers. Both founders left their undergraduate programs to pursue their goal of building Artificial General Intelligence (AGI) with an emphasis on safety. By early 2026, the two-year-old startup, composed of a six-person team in San Francisco, secured a significant $75 million Series A funding round led by Sequoia Capital and Spark Capital, with participation from notable angel investors like Andrej Karpathy. This investment valued the company at approximately $500 million.

Standard Intelligence's first major output is FDM-1, a foundation model trained on an 11-million-hour dataset of computer usage videos. The model can perform a range of complex activities, such as designing components in CAD software, identifying bugs in applications, and even learning to drive a vehicle through a web interface after a short period of fine-tuning. A core component of their technology is a highly efficient video encoder and a proprietary 30-petabyte data storage cluster built cost-effectively in-house. The company's primary objective is to use its capital to scale computation and research, aiming to create general learning models that can explore and acquire skills in new digital environments, pushing beyond the boundaries of their initial training data.

Keywords: artificial general intelligence, AGI, foundation models, computer agents, video pre-training, computer vision, agentic AI, FDM-1, Sequoia Capital, Spark Capital, Galen Mead, Devansh Pandey, human-computer interaction, AI safety, machine learning, knowledge work automation, pixel-based learning, computer action model